Low Cost Wearable AI Knee Rehabilitation System: Real-Time Motion Correction and Adaptive Training Based on Multimodal Sensing Fusion

The success of knee rehabilitation depends heavily on the correct execution of exercises. Currently, most treatments require supervision by physical therapists, which can be expensive. When exercises are performed without a therapist, incorrect movements are often used, increasing the risk of re-injury. To address this, a low-cost, wearable smart knee rehabilitation system is introduced in this study. A combination of multimodal sensors is employed to provide real-time motion correction and adaptive training. By integrating IMU data, visual pose estimation, and muscle signals (EMG), three-dimensional knee angles are accurately measured. Common mistakes such as knee collapse, twisting, or insufficient bending can be detected. Instant feedback is delivered through a display interface, guiding users to correct their posture and thereby improving exercise quality while reducing injury risk. Additionally, the system incorporates a smart training feature. Surface EMG (sEMG) is used to monitor muscle fatigue and movement accuracy. When fatigue is detected, the workout is automatically adjusted—by reducing speed or decreasing the number of repetitions. This approach reduces the dependence on expensive clinical equipment. The integration of multimodal sensor data with an adaptive system makes home rehabilitation more scientific, accurate, and user-friendly. Finally, this research provides a low-cost, high-precision real-time tracking prototype for smart home monitoring.